ModuleNotFoundError: No module named `pytz` when importing `sagemaker.workflow.pipeline`

[maxiuboldi]

• PySDK V2 (2.x)
• PySDK V3 (3.x)

Describe the bug

When importing the SageMaker Pipelines module the runtime raises:

ModuleNotFoundError: No module named 'pytz'

The SDK imports pytz directly from a pipeline module but does not declare pytz in its own install-time dependencies. If dependency resolution installs pandas>=3.x (which no longer pulls pytz), pytz may be absent and the import fails.

To reproduce

Create a fresh environment and run the following steps (complete and runnable):

1. Create and activate a virtualenv, then install the SDK and pandas 3.x:

python -m venv venv
source venv/bin/activate
python -m pip install --upgrade pip
pip install "sagemaker==2.257.3" "pandas==3.0.0"

2. Create reproduce.py with this exact content:

# reproduce.py
from sagemaker.workflow.pipeline import Pipeline

print("Imported Pipeline OK")

3. Run:

python reproduce.py

Observed result: import fails with a traceback like:

Traceback (most recent call last):
  File "reproduce.py", line 1, in <module>
    from sagemaker.workflow.pipeline import Pipeline
  File ".../site-packages/sagemaker/workflow/pipeline.py", line XX, in <module>
    import pytz
ModuleNotFoundError: No module named 'pytz'

Expected behavior

Importing public SDK modules (e.g., sagemaker.workflow.pipeline) should not raise ModuleNotFoundError for direct imports used by the SDK. Installing the SDK alone should provide all runtime dependencies required by its modules.

Screenshots or logs

See the traceback above produced by the reproduction steps.

System information

• SageMaker Python SDK version: sagemaker==2.257.3
• Framework name or algorithm: N/A (issue occurs at SDK import time)
• Framework version: N/A
• Python version: Python 3.12
• CPU or GPU: CPU
• Custom Docker image (Y/N): N

Additional context

• Root cause: Historically pandas<3 depended on pytz, so pytz was often pulled transitively. With pandas>=3 that transitive dependency is removed, exposing the SDK's missing direct dependency on pytz.